Face Generation¶

In this project, you'll use generative adversarial networks to generate new images of faces.

Get the Data¶

You'll be using two datasets in this project:

MNIST
CelebA

Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.

If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Found mnist Data
Found celeba Data

Explore the Data¶

MNIST¶

As you're aware, the MNIST dataset contains images of handwritten digits. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x7f7cf3508438>

CelebA¶

The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x7f7cf0316ba8>

Preprocess the Data¶

Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.

The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).

Build the Neural Network¶

You'll build the components necessary to build a GANs by implementing the following functions below:

model_inputs
discriminator
generator
model_loss
model_opt
train

Check the Version of TensorFlow and Access to GPU¶

This will check to make sure you have the correct version of TensorFlow and access to a GPU

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.0.1
Default GPU Device: /gpu:0

Input¶

Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:

Real input images placeholder with rank 4 using image_width, image_height, and image_channels.
Z input placeholder with rank 2 using z_dim.
Learning rate placeholder with rank 0.

Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    # TODO: Implement Function
    real_input = tf.placeholder(dtype=tf.float32, shape=[None, 28, 28, image_channels], name="real_input_images_ph")
    z = tf.placeholder(dtype=tf.float32, shape=[None, z_dim], name="generator_noise_input_ph")
    learning_rate = tf.placeholder(dtype=tf.float32, name="learning_rate_ph")
    return real_input, z, learning_rate


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

def dense(x, out_units):
    return tf.layers.dense(x, out_units, activation=None, use_bias=False, 
                           kernel_initializer=tf.random_normal_initializer(mean=0.0, stddev=0.02))

def conv2d(x, output_space):
    return tf.layers.conv2d(x, output_space, kernel_size=5, strides=2, padding="same", activation=None, 
                            use_bias=False, kernel_initializer=tf.random_normal_initializer(mean=0.0, stddev=0.02))

def leaky_relu(x, alpha=0.2):
     # non-linear activation function
    return tf.maximum(alpha * x, x)

Discriminator¶

Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variabes in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the generator, tensor logits of the generator).

def discriminator(images, reuse=False):
    """
    Create the discriminator network
    :param image: Tensor of input image(s) - shape = (None,28,28,3)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """

    with tf.variable_scope("discriminator", reuse=reuse):
        # Use LeakyReLU activation in the discriminator for all layers but the input layer.
        # convolution > Leaky RELUs, we do not use batch_norm on the first layer of the discriminator
        conv1 = conv2d(images, 64)
        conv1 = leaky_relu(conv1)
        
        # convolution > batch normalization > Leaky RELUs
        conv2 = conv2d(conv1, 128)
        conv2 = tf.layers.batch_normalization(conv2, training=True)
        conv2 = leaky_relu(conv2)
        
        conv3 = conv2d(conv2, 256)
        conv3 = tf.layers.batch_normalization(conv3, training=True)
        conv3 = leaky_relu(conv3) # non-linear activation function
        
        flatten = tf.contrib.layers.flatten(conv3)
        
        logits = dense(flatten, 1)
        output = tf.nn.sigmoid(logits)
        
    return output, logits


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

Generator¶

Implement generator to generate an image using z. This function should be able to reuse the variabes in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.

def generator(z, out_channel_dim, is_train=True):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    # Use ReLU activation in generator for all layers except for the output, which uses Tanh
    with tf.variable_scope('generator', reuse= not is_train):
        x = dense(z, 2*2*512)
        x = tf.reshape(x, [-1,2,2,512])
        x = tf.layers.batch_normalization(x, training=is_train)
        x = tf.nn.relu(x) 
        # shape=(?, 2, 2, 512)
        
        t_conv1 = tf.layers.conv2d_transpose(x, 256, 5, strides=2, padding="valid", activation=None, use_bias=False)
        t_conv1 = tf.layers.batch_normalization(t_conv1, training=is_train)
        t_conv1 = tf.nn.relu(t_conv1)
        # shape=(?, 7, 7, 256)
        
        t_conv2 = tf.layers.conv2d_transpose(t_conv1, 128, 5, strides=2, padding="same", activation=None, use_bias=False)
        t_conv2 = tf.layers.batch_normalization(t_conv2, training=is_train)
        t_conv2 = tf.nn.relu(t_conv2)
        # shape=(?, 14, 14, 128)
        
        # we do not apply batch normalization to the last layer of the generator
        # also, we apply tanh instead of relu to the output layer of the generator
        output = tf.layers.conv2d_transpose(t_conv2, out_channel_dim, 5, strides=2, padding="same", activation=tf.nn.tanh, use_bias=False)

        return output


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

Loss¶

Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    smooth = 0.1
    generated_image = generator(input_z, out_channel_dim, is_train=True)
    
    # we build 2 discriminator nets, one for real data one for fake data (comming from the generator)
    d_model_real, d_logits_real = discriminator(input_real, reuse=False)
    
    # reuse the variables here so that a new set of variables will not be created
    d_model_fake, d_logits_fake = discriminator(generated_image, reuse=True) 
    
    # since these are all real images, wee want the labels to be all ones, with label smoothing
    d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real, 
                                                                         labels=tf.ones_like(d_model_real) * (1 - smooth)))
    
    # Since we want the discriminator to output 1 for real images and 0 for fake images, the logits here are all zeros
    d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, 
                                                                         labels=tf.zeros_like(d_model_fake)))
    
    # for the generator loss, we use the fake logits from the discriminator with labels all ones
    generator_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, 
                                                                            labels=tf.ones_like(d_model_fake) * (1 - smooth)))
    
    # for the discriminator, the total loss is the sum of the losses for the real and fake images
    discriminator_loss = d_loss_real + d_loss_fake
    
    return discriminator_loss, generator_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

Optimization¶

Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    all_vars = tf.trainable_variables()
    g_vars = [var for var in all_vars if var.name.startswith("generator")]
    d_vars = [var for var in all_vars if var.name.startswith("discriminator")]
    
    # If we don't include the update ops as dependencies on the train step, the 
    # tf.layers.batch_normalization layers won't update their population statistics,
    # which will cause the model to fail at inference time
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
        discriminator_train_op = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
        generator_train_op = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)

    return discriminator_train_op, generator_train_op


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

Neural Network Training¶

Show Output¶

Use this function to show the current output of the generator during training. It will help you determine how well the GANs is training.

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

Train¶

Implement train to build and train the GANs. Use the following functions you implemented:

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """

    image_width, image_height, image_channels = data_shape[1], data_shape[2], data_shape[3]
    
    # get the inputs and learning rate placeholders
    real_input, z, lr = model_inputs(image_width, image_height, image_channels, z_dim)
    
    # get the model loss operations
    d_loss, g_loss = model_loss(real_input, z, image_channels)
    
    # get the training operations
    d_train_op, g_train_op = model_opt(d_loss, g_loss, lr, beta1)

    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            for batch_counter, batch_images in enumerate(get_batches(batch_size)):
                # normalize the batch images to be between -1 and 1
                batch_images *= 2
                
                # generate a random noise vector for feeding the generator
                batch_z = np.random.uniform(-1.0, 1.0, size=(batch_size, z_dim))
                
                # run network optimizers
                _, d_train_loss = sess.run([d_train_op, d_loss], feed_dict={real_input: batch_images, z: batch_z, lr: learning_rate})
                
                # considere to run the generator optimization operation twice to make sure the discriminator loss will NOT go to zero
                _, g_train_loss = sess.run([g_train_op, g_loss], feed_dict={real_input: batch_images, z: batch_z, lr: learning_rate})
                # _, g_train_loss = sess.run([g_train_op, g_loss], feed_dict={real_input: batch_images, z: batch_z, lr: learning_rate})

                # display the generator and discriminator loss during training
                if batch_counter % 10 == 0:
                    print("Epoch {}/{}...".format(epoch_i+1, epoch_count),
                          "Discriminator Loss: {:.4f}...".format(d_train_loss),
                          "Generator Loss: {:.4f}".format(g_train_loss))    
                
                # display some of the generator output images during training
                if batch_counter % 100 == 0:    
                    show_generator_output(sess, 20, z, image_channels, data_image_mode)

MNIST¶

Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.

batch_size = 128
z_dim = 100
learning_rate = 0.0002
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 10

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Epoch 1/10... Discriminator Loss: 1.8289... Generator Loss: 3.0356

Epoch 1/10... Discriminator Loss: 0.8060... Generator Loss: 6.1116
Epoch 1/10... Discriminator Loss: 0.6673... Generator Loss: 6.1384
Epoch 1/10... Discriminator Loss: 0.6128... Generator Loss: 5.5492
Epoch 1/10... Discriminator Loss: 0.6858... Generator Loss: 5.6026
Epoch 1/10... Discriminator Loss: 0.5178... Generator Loss: 4.0370
Epoch 1/10... Discriminator Loss: 0.4567... Generator Loss: 6.6727
Epoch 1/10... Discriminator Loss: 0.4849... Generator Loss: 4.9564
Epoch 1/10... Discriminator Loss: 0.6156... Generator Loss: 5.9406
Epoch 1/10... Discriminator Loss: 0.4148... Generator Loss: 3.7506
Epoch 1/10... Discriminator Loss: 0.5109... Generator Loss: 5.0279

Epoch 1/10... Discriminator Loss: 0.4563... Generator Loss: 4.7063
Epoch 1/10... Discriminator Loss: 0.6208... Generator Loss: 4.5739
Epoch 1/10... Discriminator Loss: 0.4730... Generator Loss: 4.2600
Epoch 1/10... Discriminator Loss: 0.4763... Generator Loss: 4.7375
Epoch 1/10... Discriminator Loss: 0.4795... Generator Loss: 3.4973
Epoch 1/10... Discriminator Loss: 0.5379... Generator Loss: 4.0865
Epoch 1/10... Discriminator Loss: 0.4704... Generator Loss: 3.8983
Epoch 1/10... Discriminator Loss: 0.4133... Generator Loss: 4.0757
Epoch 1/10... Discriminator Loss: 0.4258... Generator Loss: 4.7663
Epoch 1/10... Discriminator Loss: 0.4109... Generator Loss: 4.4972

Epoch 1/10... Discriminator Loss: 0.4417... Generator Loss: 4.8793
Epoch 1/10... Discriminator Loss: 0.4614... Generator Loss: 4.7986
Epoch 1/10... Discriminator Loss: 0.4111... Generator Loss: 4.1551
Epoch 1/10... Discriminator Loss: 0.4795... Generator Loss: 3.7431
Epoch 1/10... Discriminator Loss: 0.4800... Generator Loss: 3.4943
Epoch 1/10... Discriminator Loss: 0.4305... Generator Loss: 4.2943
Epoch 1/10... Discriminator Loss: 0.4544... Generator Loss: 4.1131
Epoch 1/10... Discriminator Loss: 0.4487... Generator Loss: 3.8108
Epoch 1/10... Discriminator Loss: 0.4224... Generator Loss: 3.6259
Epoch 1/10... Discriminator Loss: 0.4839... Generator Loss: 4.2630

Epoch 1/10... Discriminator Loss: 0.4615... Generator Loss: 4.0781
Epoch 1/10... Discriminator Loss: 0.5885... Generator Loss: 4.9578
Epoch 1/10... Discriminator Loss: 0.4881... Generator Loss: 3.6097
Epoch 1/10... Discriminator Loss: 0.5176... Generator Loss: 3.1193
Epoch 1/10... Discriminator Loss: 0.4344... Generator Loss: 4.4724
Epoch 1/10... Discriminator Loss: 0.4406... Generator Loss: 3.9667
Epoch 1/10... Discriminator Loss: 0.4413... Generator Loss: 4.0567
Epoch 1/10... Discriminator Loss: 0.6181... Generator Loss: 3.6221
Epoch 1/10... Discriminator Loss: 0.4341... Generator Loss: 4.1187
Epoch 1/10... Discriminator Loss: 0.4791... Generator Loss: 3.6221

Epoch 1/10... Discriminator Loss: 0.4108... Generator Loss: 4.2957
Epoch 1/10... Discriminator Loss: 0.4706... Generator Loss: 3.4861
Epoch 1/10... Discriminator Loss: 0.4896... Generator Loss: 3.7276
Epoch 1/10... Discriminator Loss: 0.4227... Generator Loss: 4.2172
Epoch 1/10... Discriminator Loss: 0.4652... Generator Loss: 3.0668
Epoch 1/10... Discriminator Loss: 0.4616... Generator Loss: 3.6496
Epoch 2/10... Discriminator Loss: 0.4578... Generator Loss: 3.8317

Epoch 2/10... Discriminator Loss: 0.5372... Generator Loss: 4.4035
Epoch 2/10... Discriminator Loss: 0.4436... Generator Loss: 4.3517
Epoch 2/10... Discriminator Loss: 0.4463... Generator Loss: 3.5384
Epoch 2/10... Discriminator Loss: 0.8113... Generator Loss: 2.5930
Epoch 2/10... Discriminator Loss: 0.5837... Generator Loss: 3.4538
Epoch 2/10... Discriminator Loss: 0.5603... Generator Loss: 3.5020
Epoch 2/10... Discriminator Loss: 0.4758... Generator Loss: 4.3786
Epoch 2/10... Discriminator Loss: 0.4847... Generator Loss: 3.7040
Epoch 2/10... Discriminator Loss: 0.5051... Generator Loss: 5.0259
Epoch 2/10... Discriminator Loss: 0.7279... Generator Loss: 5.9834

Epoch 2/10... Discriminator Loss: 0.6510... Generator Loss: 5.3239
Epoch 2/10... Discriminator Loss: 0.5023... Generator Loss: 4.2420
Epoch 2/10... Discriminator Loss: 0.7517... Generator Loss: 1.7330
Epoch 2/10... Discriminator Loss: 0.6974... Generator Loss: 5.6693
Epoch 2/10... Discriminator Loss: 0.5362... Generator Loss: 2.7835
Epoch 2/10... Discriminator Loss: 0.4955... Generator Loss: 3.4329
Epoch 2/10... Discriminator Loss: 0.5112... Generator Loss: 3.4043
Epoch 2/10... Discriminator Loss: 0.6366... Generator Loss: 3.7665
Epoch 2/10... Discriminator Loss: 0.5611... Generator Loss: 4.0904
Epoch 2/10... Discriminator Loss: 0.5937... Generator Loss: 2.3892

Epoch 2/10... Discriminator Loss: 0.5692... Generator Loss: 3.9585
Epoch 2/10... Discriminator Loss: 0.7773... Generator Loss: 2.9774
Epoch 2/10... Discriminator Loss: 0.5701... Generator Loss: 3.7870
Epoch 2/10... Discriminator Loss: 0.5240... Generator Loss: 3.0618
Epoch 2/10... Discriminator Loss: 0.5388... Generator Loss: 4.0750
Epoch 2/10... Discriminator Loss: 0.7905... Generator Loss: 5.8368
Epoch 2/10... Discriminator Loss: 0.6079... Generator Loss: 3.8112
Epoch 2/10... Discriminator Loss: 0.5874... Generator Loss: 3.1576
Epoch 2/10... Discriminator Loss: 0.5973... Generator Loss: 2.8620
Epoch 2/10... Discriminator Loss: 0.5285... Generator Loss: 3.3286

Epoch 2/10... Discriminator Loss: 0.5368... Generator Loss: 2.6805
Epoch 2/10... Discriminator Loss: 0.6537... Generator Loss: 2.9062
Epoch 2/10... Discriminator Loss: 0.7124... Generator Loss: 2.5623
Epoch 2/10... Discriminator Loss: 0.5728... Generator Loss: 2.9694
Epoch 2/10... Discriminator Loss: 0.5513... Generator Loss: 2.9138
Epoch 2/10... Discriminator Loss: 0.7985... Generator Loss: 1.6357
Epoch 2/10... Discriminator Loss: 0.6219... Generator Loss: 2.6957
Epoch 2/10... Discriminator Loss: 0.6320... Generator Loss: 2.3854
Epoch 2/10... Discriminator Loss: 0.9555... Generator Loss: 3.7651
Epoch 2/10... Discriminator Loss: 0.6796... Generator Loss: 3.3701

CelebA¶

Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.

batch_size = 128
z_dim = 100
learning_rate = 0.0002
beta1 = 0.5



"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 25

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

Epoch 1/25... Discriminator Loss: 1.1657... Generator Loss: 3.9644

Epoch 1/25... Discriminator Loss: 1.3379... Generator Loss: 5.8474
Epoch 1/25... Discriminator Loss: 0.9990... Generator Loss: 5.9690
Epoch 1/25... Discriminator Loss: 0.8662... Generator Loss: 5.3855
Epoch 1/25... Discriminator Loss: 0.8937... Generator Loss: 6.3943
Epoch 1/25... Discriminator Loss: 0.6907... Generator Loss: 6.0905
Epoch 1/25... Discriminator Loss: 0.6348... Generator Loss: 4.8645
Epoch 1/25... Discriminator Loss: 0.5983... Generator Loss: 5.0067
Epoch 1/25... Discriminator Loss: 0.5875... Generator Loss: 5.6049
Epoch 1/25... Discriminator Loss: 0.6614... Generator Loss: 4.2201
Epoch 1/25... Discriminator Loss: 0.5630... Generator Loss: 3.7237

Epoch 1/25... Discriminator Loss: 0.6087... Generator Loss: 5.9108
Epoch 1/25... Discriminator Loss: 0.7466... Generator Loss: 3.7985
Epoch 1/25... Discriminator Loss: 0.6074... Generator Loss: 5.2280
Epoch 1/25... Discriminator Loss: 0.5287... Generator Loss: 3.5727
Epoch 1/25... Discriminator Loss: 0.6090... Generator Loss: 4.5823
Epoch 1/25... Discriminator Loss: 0.4399... Generator Loss: 4.4540
Epoch 1/25... Discriminator Loss: 0.5092... Generator Loss: 3.9270
Epoch 1/25... Discriminator Loss: 0.5779... Generator Loss: 3.2662
Epoch 1/25... Discriminator Loss: 1.0096... Generator Loss: 3.9561
Epoch 1/25... Discriminator Loss: 0.8797... Generator Loss: 2.6920

Epoch 1/25... Discriminator Loss: 0.8382... Generator Loss: 2.9870
Epoch 1/25... Discriminator Loss: 0.6978... Generator Loss: 3.4041
Epoch 1/25... Discriminator Loss: 0.7078... Generator Loss: 3.2332
Epoch 1/25... Discriminator Loss: 0.4810... Generator Loss: 2.5158
Epoch 1/25... Discriminator Loss: 0.6033... Generator Loss: 3.7952
Epoch 1/25... Discriminator Loss: 0.8902... Generator Loss: 3.3311
Epoch 1/25... Discriminator Loss: 0.6005... Generator Loss: 2.8194
Epoch 1/25... Discriminator Loss: 1.1136... Generator Loss: 1.8822
Epoch 1/25... Discriminator Loss: 0.6887... Generator Loss: 3.5238
Epoch 1/25... Discriminator Loss: 0.9347... Generator Loss: 4.6982

Epoch 1/25... Discriminator Loss: 0.6289... Generator Loss: 3.6589
Epoch 1/25... Discriminator Loss: 0.9483... Generator Loss: 2.1025
Epoch 1/25... Discriminator Loss: 0.5876... Generator Loss: 3.9831
Epoch 1/25... Discriminator Loss: 0.6342... Generator Loss: 3.4590
Epoch 1/25... Discriminator Loss: 0.7914... Generator Loss: 2.8683
Epoch 1/25... Discriminator Loss: 0.6169... Generator Loss: 2.8483
Epoch 1/25... Discriminator Loss: 0.7921... Generator Loss: 2.9679
Epoch 1/25... Discriminator Loss: 0.6775... Generator Loss: 2.8225
Epoch 1/25... Discriminator Loss: 0.8328... Generator Loss: 2.0033
Epoch 1/25... Discriminator Loss: 0.7892... Generator Loss: 4.2101

Epoch 1/25... Discriminator Loss: 0.6801... Generator Loss: 2.4697
Epoch 1/25... Discriminator Loss: 0.6898... Generator Loss: 3.7459
Epoch 1/25... Discriminator Loss: 0.6827... Generator Loss: 3.6549
Epoch 1/25... Discriminator Loss: 0.7508... Generator Loss: 2.9511
Epoch 1/25... Discriminator Loss: 0.5732... Generator Loss: 3.2171
Epoch 1/25... Discriminator Loss: 0.5607... Generator Loss: 2.8434
Epoch 1/25... Discriminator Loss: 0.6231... Generator Loss: 2.7395
Epoch 1/25... Discriminator Loss: 0.6230... Generator Loss: 2.9792
Epoch 1/25... Discriminator Loss: 0.7319... Generator Loss: 2.0625
Epoch 1/25... Discriminator Loss: 0.6311... Generator Loss: 2.9990

Epoch 1/25... Discriminator Loss: 0.7748... Generator Loss: 3.1531
Epoch 1/25... Discriminator Loss: 0.6380... Generator Loss: 2.7938
Epoch 1/25... Discriminator Loss: 0.6093... Generator Loss: 2.5181
Epoch 1/25... Discriminator Loss: 0.5846... Generator Loss: 3.1591
Epoch 1/25... Discriminator Loss: 0.7424... Generator Loss: 2.2398
Epoch 1/25... Discriminator Loss: 0.6974... Generator Loss: 2.0910
Epoch 1/25... Discriminator Loss: 0.6631... Generator Loss: 2.6702
Epoch 1/25... Discriminator Loss: 0.8715... Generator Loss: 2.2867
Epoch 1/25... Discriminator Loss: 0.8085... Generator Loss: 3.1749
Epoch 1/25... Discriminator Loss: 0.9240... Generator Loss: 2.8238

Epoch 1/25... Discriminator Loss: 0.9075... Generator Loss: 1.9672
Epoch 1/25... Discriminator Loss: 0.7653... Generator Loss: 2.2076
Epoch 1/25... Discriminator Loss: 0.8448... Generator Loss: 2.2607
Epoch 1/25... Discriminator Loss: 0.8433... Generator Loss: 2.3561
Epoch 1/25... Discriminator Loss: 0.8455... Generator Loss: 2.7030
Epoch 1/25... Discriminator Loss: 1.1675... Generator Loss: 2.6626
Epoch 1/25... Discriminator Loss: 0.9989... Generator Loss: 1.5260
Epoch 1/25... Discriminator Loss: 1.0289... Generator Loss: 2.8720
Epoch 1/25... Discriminator Loss: 1.2806... Generator Loss: 1.2496
Epoch 1/25... Discriminator Loss: 1.1828... Generator Loss: 2.9874

Epoch 1/25... Discriminator Loss: 1.2335... Generator Loss: 3.0176
Epoch 1/25... Discriminator Loss: 1.4544... Generator Loss: 1.6135
Epoch 1/25... Discriminator Loss: 0.7847... Generator Loss: 2.4236
Epoch 1/25... Discriminator Loss: 0.9625... Generator Loss: 2.3554
Epoch 1/25... Discriminator Loss: 1.2913... Generator Loss: 1.6392
Epoch 1/25... Discriminator Loss: 0.8497... Generator Loss: 2.6723
Epoch 1/25... Discriminator Loss: 0.8646... Generator Loss: 2.0204
Epoch 1/25... Discriminator Loss: 1.1694... Generator Loss: 1.7247
Epoch 1/25... Discriminator Loss: 1.5692... Generator Loss: 3.5271
Epoch 1/25... Discriminator Loss: 1.1369... Generator Loss: 2.1429

Epoch 1/25... Discriminator Loss: 1.1031... Generator Loss: 2.7247
Epoch 1/25... Discriminator Loss: 1.7312... Generator Loss: 0.9619
Epoch 1/25... Discriminator Loss: 0.9988... Generator Loss: 1.7202
Epoch 1/25... Discriminator Loss: 1.2003... Generator Loss: 2.0211
Epoch 1/25... Discriminator Loss: 1.4491... Generator Loss: 2.4762
Epoch 1/25... Discriminator Loss: 0.8599... Generator Loss: 2.0170
Epoch 1/25... Discriminator Loss: 0.7994... Generator Loss: 2.3288
Epoch 1/25... Discriminator Loss: 0.8040... Generator Loss: 2.2201
Epoch 1/25... Discriminator Loss: 0.9467... Generator Loss: 1.7643
Epoch 1/25... Discriminator Loss: 1.1424... Generator Loss: 3.2014

Submitting This Project¶

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.